VISOR: Agentic Visual Retrieval-Augmented Generation via Iterative Search and Over-horizon Reasoning
2026-04-10 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionArtificial Intelligence
AI summaryⓘ
The authors address problems in Visual Retrieval-Augmented Generation (VRAG) systems that struggle when important visual clues are spread across multiple pages and when searches get confused over long steps. They created VISOR, a new method that helps the system keep track of evidence across pages and carefully manage visual actions to avoid mistakes. VISOR also uses clever ways to keep the search focused and prevent overload from too much visual data. Their approach shows better performance and efficiency in tests involving multi-step visual reasoning. Overall, the authors improved how AI models understand and use complex visual documents.
Visual Retrieval-Augmented GenerationVision-Language ModelsMulti-step reasoningCross-page reasoningVisual actionsSearch driftEvidence SpaceReinforcement LearningDynamic TrajectoryIntent Injection
Authors
Yucheng Shen, Jiulong Wu, Jizhou Huang, Dawei Yin, Lingyong Yan, Min Cao
Abstract
Visual Retrieval-Augmented Generation (VRAG) empowers Vision-Language Models to retrieve and reason over visually rich documents. To tackle complex queries requiring multi-step reasoning, agentic VRAG systems interleave reasoning with iterative retrieval.. However, existing agentic VRAG faces two critical bottlenecks. (1) Visual Evidence Sparsity: key evidence is scattered across pages yet processed in isolation, hindering cross-page reasoning; moreover, fine-grained intra-image evidence often requires precise visual actions, whose misuse degrades retrieval quality; (2) Search Drift in Long Horizons: the accumulation of visual tokens across retrieved pages dilutes context and causes cognitive overload, leading agents to deviate from their search objective. To address these challenges, we propose VISOR (Visual Retrieval-Augmented Generation via Iterative Search and Over-horizon Reasoning), a unified single-agent framework. VISOR features a structured Evidence Space for progressive cross-page reasoning, coupled with a Visual Action Evaluation and Correction mechanism to manage visual actions. Additionally, we introduce a Dynamic Trajectory with Sliding Window and Intent Injection to mitigate search drift. They anchor the evidence space while discarding earlier raw interactions, preventing context from being overwhelmed by visual tokens. We train VISOR using a Group Relative Policy Optimization-based Reinforcement Learning (GRPO-based RL) pipeline with state masking and credit assignment tailored for dynamic context reconstruction. Extensive experiments on ViDoSeek, SlideVQA, and MMLongBench demonstrate that VISOR achieves state-of-the-art performance with superior efficiency for long-horizon visual reasoning tasks.